A stochastic branching particle method for solving non-conservative reaction–diffusion equations

Liyao Lyu, University of California, Los Angeles, 520 Portola Plaza, Los Angeles, 90095, CA, USA (.), Huan Lei, Michigan State University, 428 S Shaw Ln, Department of Statistics & Probability, Michigan State University, Wells Hall, 619 Red Cedar Road, East Lansing, 48824, MI, USA () · 2025

Read A stochastic branching particle method for solving non-conservative reaction–diffusion equations by Liyao Lyu, University of California, Los Angeles, 520 Portola Plaza, Los Angeles, 90095, CA, USA (.), Huan Lei, Michigan State University, 428 S Shaw Ln, Department of Statistics & Probability, Michigan State University, Wells Hall, 619 Red Cedar Road, East Lansing, 48824, MI, USA () (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2510-27615.txt.

Introduction

Liyao Lyu and Huan Lei propose a particle method in 2025 for equations where spread, flow, growth, and loss act at once and total mass is not kept fixed. Grid methods can strain when values blow up or form sharp clumps. Their method splits each step in two: a move-and-spread step run by random particle paths, and a growth step run by birth and death of particles.

The scheme needs no grid and never yields counts below zero. Tests cover two classic cases: one where phases split apart, and one where cells clump as they chase a chemical trail. Both patterns are matched well without fine grid tweaks. The method suits hard cases where mass shifts and peaks form.

Catalog record

CREATORLiyao Lyu, University of California, Los Angeles, 520 Portola Plaza, Los Angeles, 90095, CA, USA (.), Huan Lei, Michigan State University, 428 S Shaw Ln, Department of Statistics & Probability, Michigan State University, Wells Hall, 619 Red Cedar Road, East Lansing, 48824, MI, USA ()
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE55 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2510-27615

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A stochastic branching particle method for solving non-conservative reaction–diffusion equations Thanks: Submitted to the editors DATE. Liyao Lyu, University of California, Los Angeles, 520 Portola Plaza, Los Angeles, 90095, CA, USA (.), Huan Lei, Michigan State University, 428 S Shaw Ln, Department of Statistics & Probability, Michigan State University, Wells Hall, 619 Red Cedar Road, East Lansing, 48824, MI, USA () Abstract We propose a stochastic branching particle-based method for solving nonlinear non-conservative advection–diffusion–reaction equations. The method splits the evolution into an advection–diffusion step, based on a linearized Kolmogorov forward equation and approximated by stochastic particle transport, and a reaction step implemented through a branching birth–death process that provides a consistent temporal discretization of the underlying reaction dynamics. This construction yields a mesh-free, nonnegativity-preserving scheme that naturally accommodates non-conservative systems and remains robust in the presence of singularities or blow-up. We validate the method on two representative two-dimensional systems: the Allen–Cahn equation and the Keller–Segel chemotaxis model. In both cases, the present method accurately captures nonlinear behaviors such as phase separation and aggregation, and achieves reliable performance without the need for adaptive mesh refinement. keywords particle methods, stochastic branching, reaction-diffusion, chemotaxis, aggregation Funding. This work is supported in part by the National Science Foundation under Grant DMS-2143739 and the ACCESS program through allocation MTH210005. AMS 65M75, 65C35, 92C17 1 Introduction The advection–diffusion–reaction (ADR) type partial differential equations (PDEs) have attracted much interest in physics, biology, and engineering for modeling various phenomena relevant to chemical reactions, transport, and collective behaviors in biological systems [ 24 ] . Traditional grid-based methods, such as finite difference [ 21 ] , finite element [ 10 ] , and discontinuous Galerkin schemes [ 2 , 50 ] , have achieved great success in solving many of these problems. However, from a numerical perspective, an essential challenge in solving such PDEs lies in handling nonlinearities and singular structures in certain cases, e.g., the aggregation phenomena of the Keller–Segel (KS) model [ 27 ] and the interfacial dynamics of the Allen–Cahn (AC) model [ 1 ] . Mesh-based discretizations approximate solutions in function spaces that may not be well-suited for capturing the blow-up or highly localized behavior, often resulting in reduced accuracy or numerical instabilities, and requiring additional numerical treatments. Particle-based approaches offer a natural alternative by representing the solutions as empirical measures, which are particularly effective for problems exhibiting concentration or blow-up phenomena. Existing particle methods, such as the particle-in-cell method [ 4 , 20 ] for solving the Vlasov–Poisson equation, are often based on well-established underlying physical laws. In this setting, particle dynamics can be derived from the explicit micro-scale governing equations. In contrast, the explicit physical laws are often unknown for general non-conservative ADR type PDEs, where effective interactions may be highly nonlinear, system-specific, or even not well characterized. For example, in many biological and chemical systems such as chemotaxis, the micro-scale interactions remain poorly understood, which limits the applicability of traditional particle-based approaches. To extend particle-based methods beyond physically motivated settings, blob methods [ 11 , 5 , 23 ] and semi-Lagrangian particle schemes [ 36 , 37 , 38 ] have been developed for conservative PDEs. These approaches are based on the mean-field perspective, in which the empirical measures of interacting particle systems converge to the solution of the original PDE. This mathematical framework establishes a rigorous link between the particle dynamics and the continuum equations and does not rely on explicit microscale physical laws. Despite these advances, there is still no unified particle-based framework for systematically discretizing nonlinear and non-conservative PDEs, especially in the presence of singularities, sharp interface or blow-up phenomena. Machine learning-based numerical PDE algorithms, such as Deep Galerkin Methods [ 40 ] , Deep Ritz method [ 51 ] , and Mixed Residual method [ 30 ] , have attracted significant attention in recent years. By leveraging the strong representability of neural networks, these methods are particularly promising for solving high-dimensional problems. In the context of particle-based approaches, coupling neural networks with the Feynman-Kac formula [ 16 , 26 , 31 , 3 ] or backward stochastic differential equations (SDEs) [ 18 , 17 , 25 ] has also emerged as a powerful strategy for solving a broad class of PDEs. Nevertheless, the approximation capability of neural networks is fundamentally grounded in the universal approximation theorem that targets continuous solutions on compact sets. As a result, neural-network–based solvers may face intrinsic difficulties when dealing with nonlinear, non-conservative PDEs that develop singularities or blow-up phenomena, where discontinuities or concentration effects play a central role. In this work, we propose a stochastic branching particle method to address these challenges. At each time step, the original PDE is locally linearized and approximated by the Kolmogorov forward equation, which establishes a direct connection between the PDE and the stochastic particle dynamics. Unlike approaches that rely on explicit physical laws or mean-field limit arguments, this framework establishes a systematic particle-based discretization applicable to general nonlinear and non-conservative PDEs. A key feature of the method is a particle-level birth–death mechanism, introduced to represent non-conservative reaction terms by allowing the number of particles to evolve consistently with the underlying reaction dynamics. The only step that invokes a continuity assumption is the interpolation of the empirical measure, which can be replaced by blob- or kernel-based representations. Consequently, the developed method does not fundamentally rely on continuity and is well-suited for problems with singularities or blow-up phenomena. From a numerical perspective, the proposed method is broadly compatible with existing stochastic interpretations of reaction–diffusion systems. For example, in the case of chemotaxis, it aligns with the mean-field derivation of the KS equation from moderately interacting particles developed by Stevens [ 42 ] . However, our formulation is not restricted to this setting and applies generally to nonlinear ADR systems with non-conservative terms. To illustrate the effectiveness of the proposed method, we consider two representative non-conservative reaction–diffusion systems: the AC equation for the phase separation with interfacial motion and nonlinear reactions, and the KS chemotaxis model for the aggregation-driven blow-up dynamics [ 27 , 33 ] . While the AC equation serves to demonstrate the generality of our method, the KS model provides a more challenging test problem and is the primary focus of the present study. In particular, we consider the two-dimensional KS chemotaxis model as the theme problem, which takes the form \displaystyle\partial_{t}u-\nabla\cdot(\nabla u-\chi(u)\nabla v) \displaystyle=f_{u}(u,v), \displaystyle\mathbf{x}\in\mathbb{T}^{2},t\in(0,T), (1) \displaystyle\partial_{t}v-\Delta v \displaystyle=f_{v}(u,v), \displaystyle\mathbf{x}\in\mathbb{T}^{2},t\in(0,T), where u and v represent the cell density and chemical concentration, respectively. \chi represents the chemotactic sensitivity, and f_{u} and f_{v} represent the production or decay rate. For simplicity, we focus on periodic boundary conditions or unbounded domains, though the framework can be extended to more general boundary settings in future research. In two dimensions, it is well understood that both parabolic–elliptic and fully parabolic cases can exhibit finite-time blow-up to a Dirac-delta profile when the initial mass is sufficiently large [ 19 , 35 ] . Furthermore, in the three-dimensional parabolic–elliptic case, an alternative self-similar blow-up profile of the form C(T-t+|\mathbf{x}|^{2})^{-1} has also been identified in the literature [ 15 , 41 ] . On the numerical side, many notable methods have been developed for the KS systems. The finite-volume method [ 14 ,

Read the complete text